I Have ADHD is a free, open source Claude Code plugin that rewrites every AI reply to lead with the next action instead of burying it under a paragraph of preamble, and it is currently the hottest repository on GitHub, adding roughly 3,900 stars in a single day on top of nearly 38,000 total. Setup is two commands and takes under two minutes: install the plugin, restart your session, and type one slash command to turn it on.

  • Install is two commands: claude plugin marketplace add then claude plugin install, no signup, no API key, no account.
  • It is off until you invoke it. Type /i-have-adhd to turn the ruleset on for that session, or flip a flag file to make it load automatically at every session start.
  • The whole ruleset is 10 short rules in one editable file (SKILL.md), so you can fork it and rewrite the rules to match your own reading style.
  • It is not Claude Code only: the same repository ships install instructions for 15 different agent harnesses, from Codex and Gemini CLI to GitHub Copilot and Cursor.

The exact steps, start to finish

  1. Step 1. Confirm the prerequisites.
    # check the Claude Code CLI and git are both installed
claude --version
git --version
  2. Step 2. Add the plugin marketplace.
    claude plugin marketplace add ayghri/i-have-adhd
  3. Step 3. Install the plugin.
    claude plugin install i-have-adhd@i-have-adhd
  4. Step 4. Verify it installed.
    claude plugin list
    Confirm i-have-adhd appears in the output.
  5. Step 5. Restart Claude Code and invoke the skill. Start a new session (the plugin index is only read at startup), then type /i-have-adhd inside that session to turn the ruleset on.
  6. Step 6 (optional). Make it always-on instead of per-session.
    # loads the full ruleset from message one, every session
touch ~/.claude/.i-have-adhd-always
    Delete that file later to go back to on-demand.
  7. Step 7. Put it to work. With the skill on, ask a real question, for example "explain how to create an empty Git repository in a new folder." A working install replies with the command first, numbered steps if there is more than one, and no "Hope this helps!" at the end.
Typical AI reply versus a reply with I Have ADHD onA typical AI reply opens with a compliment, explains before it answers, buries the command mid paragraph, and closes with a stock sign off. With I Have ADHD on, the same model leads with the exact command, numbers each step, keeps one action per line, and ends on a single next step. TYPICAL AI REPLY Opens with "Great question!" Explains before it answers Buries the command mid-reply Closes with "Hope this helps!" WITH I HAVE ADHD ON Leads with the exact command Numbers each step One action per line Ends on one next step Same model, same question, only the shape of the reply changes genztech.blog
Fig 1 The plugin does not change the model, it changes the shape of the model's reply: action first, numbered, and short.

What is I Have ADHD and why is it trending?

I Have ADHD is an MIT-licensed Claude Code plugin from developer ayghri, created in May 2026 and now sitting near 38,000 stars with 2,187 forks. It is not a model, a wrapper, or a fine-tune: it is a single Markdown file of ten output rules (lead with the next action, number multi-step tasks, cap lists at five items, no preamble, no recap, no closers) that gets loaded into the assistant's context when you invoke it. The pitch is specific: coding assistants default to a chatty, hedge-everything register that reads fine as prose but is genuinely hard to act on if your attention keeps sliding off the paragraph before the command arrives. The README's before-and-after comparison makes the target obvious: a wall of text about "the middleware, the token verification, and the cookie handling" on one side, three numbered lines ending in a test command on the other. That is a narrow, concrete problem, which is probably why it spread this fast. It also credits its structure to a real clinical source, The Adult ADHD Tool Kit by Ramsay and Rostain, adapted for how a language model should respond rather than how a person should plan a day.

RelatedOpenMontage Setup: Turn Your AI Agent Into a Video Studio

How do you install I Have ADHD in Claude Code?

The whole install is the two commands in the checklist above: claude plugin marketplace add ayghri/i-have-adhd registers the repository as a plugin source, then claude plugin install i-have-adhd@i-have-adhd pulls the plugin itself. Neither command touches your project files or asks for credentials, since the plugin only ever adds text to the system prompt when you invoke it. After installing, restart the CLI (Claude Code reads the plugin index at startup, so a mid-session install will not show up in autocomplete) and run claude plugin list to confirm it landed. From there, typing /i-have-adhd turns the ruleset on for that session only; typing "stop adhd mode" or "normal mode" turns it back off without restarting anything. Updating later is claude plugin marketplace update i-have-adhd, and removing it is the marketplace-add sequence in reverse: claude plugin uninstall i-have-adhd then claude plugin marketplace remove i-have-adhd.

Does it work outside Claude Code?

Yes, and this is the part that explains the fork count. The same repository ships tested, harness-specific install steps for Codex (codex plugin marketplace add), Gemini CLI (a custom command or an always-on extension), GitHub Copilot and any Agent-Skills-compatible tool via npx skills add ayghri/i-have-adhd, OpenCode, Pi, Qwen Code, Zed, Kimi Code CLI, and a few more, fourteen harnesses in total plus Claude Code. The mechanism differs per tool (a plugin manifest here, a `SessionStart` hook there, a native Agent Skills folder elsewhere) but the underlying artifact is the same `SKILL.md`, so the behavior you get in Claude Code is the behavior you get everywhere else it is wired up. That is unusually thorough maintenance for a project four months old, and it is the reason this reads as a real piece of tooling infrastructure rather than a one-off prompt trick.

How do you turn it on for every session instead of typing a command?

On demand is the default in Claude Code, Qwen Code, and Codex, meaning nothing changes until you explicitly ask for it. If retyping /i-have-adhd every session is itself the friction, the flag file in step 6 fixes that: touch ~/.claude/.i-have-adhd-always makes a SessionStart hook load the full ruleset automatically, every time, with no slash command needed. Delete the file to go back to opt-in. The README notes one interaction worth knowing: a saved "stop adhd mode" choice for the current session always overrides the always-on flag, so you can dial it back for one conversation without touching the flag file at all.

RelatedVoiceStudio Setup: Clone Voices Locally, No Cloud

What are the gotchas before you rely on it?

Three things worth knowing before you make this your default. First, the rules only apply when the harness actually honors `disable-model-invocation`, which Claude Code, Qwen Code, Codex, and GitHub Copilot do (Copilot's behavior was specifically verified in the project's own pull request #60); other harnesses may read every installed skill's description at startup and decide for themselves whether to activate it, so behavior can vary by tool. Second, this changes tone and structure, not judgment: the rules explicitly carve out exceptions for full explanations when you ask for them and for confirmation before destructive actions, so it will not silently skip a safety check to stay terse. Third, it is genuinely just a system-prompt addition wrapped in solid packaging: there is no model change and no guarantee a provider will not add stricter default formatting that makes this redundant. None of that is a reason to skip it; it is the difference between installing it once and understanding what it actually does before it reshapes a hundred replies a day.

TraitI Have ADHDHand-written system promptDefault assistant output
Setup timeUnder 2 minutesWrite and tune it yourselfNone needed
Editable rulesetYes, fork one Markdown fileYes, but you own it aloneNo
Works across tools15 harnesses, one repoCopy-paste per tool by handN/A
On by defaultNo, explicit invoke or a flagDepends where you put itAlways verbose
PriceFree, MIT licensedFree, your time onlyFree
What to watch · 2026
  • Harness drift. Fifteen install paths is a lot of surface area to keep working as each agent CLI ships breaking changes; watch whether the 47 open issues stay dominated by harness-specific bugs or by the core rules themselves.
  • Implicit-activation harnesses. Tools that do not honor `disable-model-invocation` could start auto-applying the ruleset without a user asking, which would undercut the project's own "installed, not invoked" design goal.
  • Whether providers ship this by default. If a lab bakes terser, action-first formatting into its default system prompt, a plugin like this becomes redundant for that model specifically.

Our take

Strip the branding and this is a config file: ten formatting rules that tell a model to answer first and stop performing helpfulness. That is not a new idea, plenty of people keep a personal system-prompt snippet that does something similar. What earned 38,000 stars in four months is the packaging, not the concept: one canonical `SKILL.md`, a maintained install path for fifteen different agent harnesses, explicit opt-in behavior instead of a silent takeover, and a documented escape hatch the moment you want the old chatty tone back. For anyone who finds themselves scrolling past three sentences of throat-clearing to find the actual command, two commands and a slash invocation is a genuinely low-cost trial. The real risk is not that it breaks anything, the rules are deliberately conservative about destructive actions and full explanations, it is that fifteen install surfaces are fifteen things that can quietly drift out of sync as each harness updates. Worth installing today; worth checking back on in a few months to see if the maintenance held.

Primary sources

Original analysis by GenZTech. Tool documentation: ayghri/i-have-adhd on GitHub.